This week, Cursor—a well-funded AI startup focused on software development—unveiled its latest product: Composer 2. The company’s announcement dripped with bold confidence, promising “frontier-level coding intelligence.” All the hallmarks of a major leap forward were on display.
Yet the tech sphere, ever-eager for cracks to show, quickly pounced. An X user by the name of Fynn put forth a pointed accusation: Composer 2, they claimed, was little more than “Kimi 2.5” in disguise, gussied up with some extra reinforcement learning. Kimi 2.5, for context, is an open-source project released by the Chinese firm Moonshot AI—a company with heavy backing from corporate giants like Alibaba and HongShan.
Fynn’s evidence? A snippet of code, apparently surfacing Kimi’s name as Composer 2’s backbone. His tone was unmistakably mocking: “At least change the model ID,” he shot back.
For those who’d tracked Cursor’s ascent, the assertion landed with a thud. Cursor sits on a $2.3 billion war chest after last year’s fundraising, with valuation estimates hovering near $30 billion—and reports surfacing of revenues north of $2 billion annually. The company’s public stance hadn’t hinted at outside involvement, let alone a Chinese firm’s open-source engine.
Soon enough, Lee Robinson—Cursor’s vice president overseeing developer education—addressed the swirling doubts. Publicly, he conceded: “Yep, Composer 2 started from an open-source base!” But Robinson was quick to draw a line. According to him, just around a quarter of the compute used in Composer 2’s final version originated from that initial base. The lion’s share, he insisted, came from Cursor’s own training and refinements. The benchmarks, Robinson argued, tell the story: Composer 2 and Kimi are, by his assessment, meaningfully different in performance.
Robinson was also adamant that Cursor’s actions were by the book. He referred to Kimi’s open license and later, the official Kimi account on X echoed that approval. In a post, they congratulated Cursor, clarifying that the use occurred “as part of an authorized commercial partnership” through Fireworks AI.
“We’re proud Kimi-k2.5 could serve as foundation,” the message read. Kimi’s team sounded genuinely encouraging, remarking that Cursor’s diligent retraining and extensive reinforcement learning built on the open model ecosystem they aim to support.
Still, some kept circling the original omission. Why hadn’t Cursor acknowledged Kimi up front? The possible reasons hum beneath the surface: a touch of embarrassment, perhaps; or, given the current geopolitical friction surrounding AI, unease about relying on technology born from China. Only a year ago, Silicon Valley was abuzz—if not outright panicked—after DeepSeek, another Chinese venture, launched a model that sent shockwaves through the competitive landscape.
Cursor co-founder Aman Sanger, for his part, admitted the oversight: “It was a miss to not mention the Kimi base in our blog from the start,” he wrote, adding, “We’ll fix that for the next model.”
The episode exposes the thin membrane between innovation and adaptation in the world of artificial intelligence. Behind polished launch posts and billion-dollar rounds of funding, the machinery of progress is rarely as clean or proprietary as marketing likes to pretend. Collaboration, modification, and commercial licensing—these are the real engines, humming along just below the surface.
Meanwhile, in announcements and social feeds, the cycle continues—new launches, partnerships, rivalries, and a ceaseless drumbeat of “breakthrough innovations.” But for those who pay attention, the story between the lines is just as significant: the blurred borders, the quiet partnerships, the hurried acknowledgments that follow in the wake of scrutiny.
Nothing in AI, it seems, evolves in isolation. And for every model with a new name, somewhere, the fingerprints of others quietly linger.
